[Bonmin] New stable branch and candidate for release
Pierre Bonami
pierre.bonami at gmail.com
Fri Jun 20 05:34:42 EDT 2008
Dear list,
Following Stefan (and others) gentle suggestion I think it is time to
have a new release of Bonmin.
I have created a new stable branch 0.99 and made a new release 0.99.0.
I think that the code is mostly stable (and more stable than the last
release 0.1.4). Nevertheless some more tests are needed and I think it
is safer to call it a candidate release.
There are quite a few changes in this version almost all the code has
been rewritten over the last year and a half (and some of the changes
are not yet activated). I will try to document the changes as we move
towards release 1.0. There is already a quite up-to-date option
documentation available here:
http://www.coin-or.org/Bonmin/options_list/index.html
(note that some options names may have changed)
The main major change is that the default algorithm has been changed
from B-Hyb to B-BB (i.e. outer-approximation to pure branch-and-bound).
This is mostly for two reasons. First because we did a lot of work on
B-BB and it is now a lot more efficient than it used to be. Second
because many users were trying to solve non-convex problems with the
default settings and B-Hyb gave them terribly bad results. Still on some
convex problems B-Hyb or B-QG can be much faster than B-BB and if you
find the default algorithm too slow you should try them by putting
bonmin.algorithm B-Hyb
in the bonmin.opt option file.
Another important change is that the default node selection strategy has
been changed from best-bound to dynamic. This should have the good
effect that feasible solutions are found faster but on the other hand it
may take longer to prove optimality. Again you can use best-bound using
the node_comparison option.
Cheers,
Pierre
More information about the Bonmin
mailing list